In this paper, a lower bound on the capacity of wireless ad hoc erasure networks is derived in closed form in the canonical case where $n$ nodes are uniformly and independently distributed in the unit area square. The bound holds almost surely and is asymptotically tight. We assume all nodes have fixed transmit power and hence two nodes should be within a specified distance $r_n$ of each other to overcome noise. In this context, interference determines outages, so we model each transmitter-receiver pair as an erasure channel with a broadcast constraint, i.e. each node can transmit only one signal across all its outgoing links. A lower bound of $\Theta(n r_n)$ for the capacity of this class of networks is derived. If the broadcast constraint is relaxed and each node can send distinct signals on distinct outgoing links, we show that the gain is a function of $r_n$ and the link erasure probabilities, and is at most a constant if the link erasure probabilities grow sufficiently large with $n$. Finally, the case where the erasure probabilities are themselves random variables, for example due to randomness in geometry or channels, is analyzed. We prove somewhat surprisingly that in this setting, variability in erasure probabilities increases network capacity.